Problem with MPEG-2 Audio compliance test file 'test11.mpg' 23-July-1996 ============================================================================== Concerning the test data set test11, there are often questions like: I tried to decode some test streams using reference shareware decoder and encountered a problem: test11 is encoded with invalid combination dematrix_procedure == 2, tc_alloc > 3. ISO/IEC 13818-3 says that for dematrix_procedure == '10' only tc_allocations 0, 1, 2 are valid (section 2.5.2.10). Here is an explanation by the editor of the conformance part of MPEG-2 audio (IS 13818-4) Leon van de Kerkhof tel: +31 402 734 039 Philips Sound & Vision / ASA Lab - Eindhoven fax: +31 402 737 353 Building SFJ-7.06 email: kerkhofl@ce.philips.nl P.O. Box 80 002 5600 JB Eindhoven The Netherlands -------------------------------------------------------------------------------- The bit stream 11 is indeed not compatible with 13818-3 dated 05-15-1995. At the Florence meeting it was agreed that bitstream 11 would be supplied conformant to the >revision< of 13818-3, which is in the mean time DIS. In the revision, the low-pass filter has been removed for matrix mode 2, and that made all tc_allocations possible again. Also the decoding equations have been adapted accordingly in the revision. I attach an ascii version of my input document to the Florence MPEG meeting that explains the change with respect to matrix mode 2. These changes were accepted. If this information is clear enough, it could be incorporated in the bit stream 11 description as a whole. ******************************************************************************* The following proposal was accepted in March 1996, be it with some editorial changes ******************************************************************************* INTERNATIONAL ORGANIZATION FOR STANDARDIZATION ORGANISATION INTERNATIONALE DE NORMALISATION ISO/IEC JTC1/SC29/WG11 CODING OF MOVING PICTURES AND AUDIO ISO/IEC JTC1/SC29/WG11 MPEG96/xxx March 1996 Source: Leon van de Kerkhof Title: Possible improvement to Revision of ISO/IEC 13818-3 with respect to matrix mode 2 Status Proposal 1. Introduction Recently, there has been some discussion on the implementation of matrix mode 2 in the MPEG-2 standard. This is mostly related to the Low-pass filter (table B.3), that should be exactly the same in the encoder and the decoder to obtain perfect dematrixing. Especially because of the fact that the filter should be applied before subband filtering, in the subband domain, the implementation is not as straightforward as it seems at first sight. Based on some new information which is relevant for this topic, I would like to make a change proposal to the MPEG Audio group. The current revision of ISO/IEC 13818-3 provides an opportunity to implement the proposed change. It should be checked very carefully whether existing implementations would be affected by such a change. 2. Analysis of the problem A full feature analogue surround encoder performs 4 operations: 1 7 kHz Low Pass filter on the monophonic surround signal S 2 modified Dolby B on the filtered S 3 90 degrees phaseshift on the resulting S signal 4 Matrixing : Lt =3D L + 0.7*C - 0.7*j*S Rt =3D R + 0.7*C + 0.7*j*S There are indications that, because of the good channel quality in a digital transmission of MPEG Audio coded data, the Low Pass filter (and probably also the Dolby B operation) can be omitted. Informal listening tests at Philips, where MPEG-2 coded discrete 5-channel movie material has been downmixed both with and without the Low-pass filter, indicate that this does not deteriorate the sound quality of the Pro Logic decoded signal, maybe even the contrary is true. Evaluation by movie sound experts also did not reveal negative effects. This helps in two respects. One is that the implementation of both the encoder and the decoder becomes much simpler if the filter is omitted. The second advantage would be, that now the full bandwidth surround signal is available in the compatible signal, which relaxes the current restriction on the allowable tc_allocations for matrix mode 2. This would have a positive effect on the sound quality in Fixed Bit Rate mode, or on the bit rate in Variable Bit Rate mode. The actual proposal is to remove the Low Pass filter from the IS. This would also make it possible to add more tc_allocations for matrix mode 2. It is proposed to leave the modified Dolby B processing and 90 degrees phase shift in the decoder as an option in the standard. 3. Detailed proposal The page numbers in the following refer to the official 1995 ISO/IEC 13818-3 publication. page 33, tc_allocation: Remove the sentence If dematrix_procedure =="10", only tc_allocations 0,1,2 are valid. page 34, 3/1 configuration, table B: tca = 5 T2=L, T3=R page 46, step 2: Remove this step (application of the Low Pass filter). Steps 3 and 4 should be renumbered. In what follows, Swbp has everywhere to be replaced by Sw. page 47, decoding matrices for 3/2 configuration, dematrixing procedure "10": Add decoding matrices for tc_allocations 3, 4, 5, 6 and 7: tca=3: jS = -Lo + T3 + T2 R = Ro - T2 - jS jLS = 2*jS - T4 tca=4: jS = Ro - T4 - T2 L = Lo - T2 + jS jRS = 2*jS - T3 tca=5: jLS = 0.5*(Ro - Lo + T3 - T4) jRS = jLS /* In this case stereo S retrieval is not possible. The encoder has to check whether this is permissible from a psycho-acoustics point of view */ tca=6 C = 0.5*(Ro + Lo - T2 - T3) jS = 0.5*(Ro - Lo + T3 - T2) jLS = 2*jS - T4 tca=7 C = 0.5*(Ro + Lo - T2 - T4) jS = 0.5*(Ro - Lo + T2 - T4) jRS = 2*jS - T3 page 48, decoding matrices for 3/1 configuration: Change decoding matrices for dematrixing procedure "10": tca=0 L = Lo - T2 + T3 R = Ro - T2 - T3 tca=1 C = Lo - T2 + T3 R = Ro - C - T3 tca=2 C = Ro - T2 - T3 L = Lo - C + T3 tca=3 jS = -Lo + T2 + T3 R = Ro - T2 -jS tca=4 jS = Ro - T2 - T3 L = Lo - T2 + jS tca=5 C = 0.5*(Ro + Lo - T2 - T3) jS = 0.5*(Ro - Lo + T2 - T3) Table B.3: delete completely page 67: replace Swbp by Sw page 67: delete "... and bandwidth limitation. The filter preferably to be used is described in table B.3 of Annex B. With other filters, the dematrixing will not be perfect." ******************************************************************************* _______________________________________________________________________________ Hendrik Fuchs Universitaet Hannover Institut fuer Theoretische Nachrichtentechnik und Informationsverarbeitung Appelstrasse 9A, D-30167 Hannover, Germany fuchs@tnt.uni-hannover.de phone: +49-511-762-5329 fax: +49-511-762-5333 _______________________________________________________________________________